Publication date:
Publication from:
The State Inspectorate of Architecture and Urban Development of Ukraine (DIAM) is working on implementing a new AI project to significantly improve the efficiency of the inspectorate’s operations. De Novo is providing fast and reliable cloud infrastructure for the solution.
DIAM, in collaboration with partners, is developing “Digital Inspector,” an AI-based system designed to assist with project documentation verification. Its task is to identify inconsistencies with building codes, legislation, and input data before application submission, thereby reducing errors and lightening the workload on inspectors.
The need for such a solution arose from the overburdening of the current verification model. A single construction project can be accompanied by hundreds or even thousands of pages of drawings, explanatory notes, certificates, and other materials that an inspector must process within a tight deadline. Even a formal inconsistency—a missing file, a contradiction between different sections, or a discrepancy with input data—can trigger new rounds of approval and delay the entire project.
This is precisely why DIAM decided to automate the most routine and labor-intensive stages of this work. The first pilot version of the system is expected in early summer. Itera Ukraine is the developer. To build and test the solution, the team uses De Novo’s Ukrainian cloud platform with AI accelerators. The company provided cloud computing resources with NVIDIA GPUs for six months, as well as expertise in building the solution’s architecture. Dmitry Fedorenko, De Novo’s AI/ML Business Development Director, mentors the development team, assisting with model selection and fine-tuning their cooperative operation.
Dmitry Fedorenko. Photo: De Novo Press Service
Initially, the system will primarily address three groups of common problems: incomplete document packages, contradictions between different parts of the construction project, and discrepancies with urban planning conditions, restrictions, and other input data. The main idea is to detect such risks as early as possible to increase the chances of passing approval on the first attempt. Residential construction and public buildings, where the volume of documentation is typically largest, will be the first focus area. The project’s target metric is to reduce the processing time for a single application with a complete document package from ten working days to approximately two hours.
There is another important argument in favor of AI. It will create equal digital rules of the game for all participants in the process. For the inspector who makes the decision. For the designer who prepares the documentation. For the expert who verifies it. For the developer who orders the construction. Everyone can use the same digital tool to identify inconsistencies with building codes and legislation at an early stage. This is a fundamental change in approach,
says Oleksandr Novytskyi, head of DIAM.
Oleksandr Novytskyi. Photo: DIAM Press Service.
From a technical standpoint, “Digital Inspector” is built as a combination of several large language models, including LLAMA and ChatGPT. The system must be capable of reading, recognizing, and structuring documents, including scanned ones, and then analyzing them in the context of building codes and specific requirements. A RAG (Retrieval-Augmented Generation) approach is used for this: the model is not pre-trained on the entire array of regulatory documentation, but during verification, retrieves the exact rules, restrictions, and examples needed for the specific case. The system also integrates with the Unified State Electronic System in the Construction Sector (USESCS) and Diia.
A particular emphasis of the project is data protection. Since the system works with confidential state information, infrastructure localization is critical. De Novo’s facilities are located in Ukraine, and processing takes place in a secured data center. The company holds certification of compliance with ISO/IEC 27001, 27017, 27018, 27701, PCI DSS, and КСЗІ (Comprehensive Information Protection System) standards. Data will be transmitted and stored in encrypted form, and access to documents and findings will be restricted by user roles through integration with Diia and USESCS.
In the first stage, the AI will operate in a parallel verification mode alongside the inspector, meaning the algorithm will analyze documents simultaneously with a human, and the results will be compared. If the system consistently demonstrates high-quality results, its level of autonomy may be increased. DIAM emphasizes that this is only about redistributing the workload, not replacing inspectors—the AI is intended to handle the most monotonous part of the verification, while the final decision remains with the human.
